What it does
Accelerates
Runs a fast first-pass scan so human security reviewers start from a shortlist, not a blank read-through
Agent 22 · 08 Build Enablement
Security/Privacy Scan Agent
Pre-scans AI-generated code for common security/privacy issues, ahead of mandatory human review
What it never does
Stops here
Replaces the mandatory human security/privacy review, this agent's output is a starting point, never a clearance
- Inputs
- Translation Agent's generated code
- Outputs
- Flagged issues (e.g., exposed keys, unvalidated inputs, unnecessary data capture)
- Model requirement
- Code-aware model with security-pattern retrieval; can be paired with conventional static-analysis tooling rather than LLM reasoning alone
Example prompt
"Scan this generated code for exposed credentials, unvalidated input, and unnecessary data capture. List findings by severity, do not mark anything as cleared."